In healthcare systems powered by the Internet of Things (IoT), the classification of brain tumours (BT) is crucial for the diagnosis of brain cancer disorders. The most widely used use of artificial intelligence (AI) approaches based on computer-assisted diagnostic systems (CADS) is the identification of brain cancer. However, because artificial diagnostic technologies are unreliable, clinicians are not able to identify brain tumours with them. The present study aimed to tackle the problem of inaccuracy in existing artificial diagnosis systems by introducing a robust deep learning (DL)-based method for brain tumour classification. The proposed method for classifying brain cancers is based on an upgraded hybrid Particle Swarm Optimisation (PSO) with back propagation neural network (BPNN) that leverages data from brain magnetic resonance imaging (MRI) images. When compared to the suggested method (PSOBPNN), which is based on a deep learning approach and uses an IoT healthcare environment to reliably and swiftly predict brain tumours, the newly built system shows that all existing methods perform badly. The model’s performance in classification has improved with the use of data augmentation and transfer learning approaches. The results confirmed the model’s good accuracy of 96.45% when compared to the baseline models. Based on highly predictive results, we suggest the suggested deep learning PSOBPNN model for brain cancer diagnosis in IoT healthcare systems. Current methods based on CNN, ANN, and KNN algorithms are utilised to categorise MRI brain disorders. A deep learning classification algorithm is created using MATLAB 2013a in order to evaluate the efficacy of the proposed method.